MAXIMAL CHAINS IN AND ULTRAPOWERS OF THE INTEGERS

SAHARON SHELAH AND JURIS STEPRANS

Abstract. Various questions posed by P. Nyikos concerning ultralters on and chains in the partial order (, < ) are answered. The main tool is the oracle chain condition and variations of it. Keywords: ultralter, ultraproduct, oracle chain condition, Cohen real

1. Introduction In [5] various axioms related to maximal chains in ultrapowers of the integers were classied and studied. The purpose of the present paper is to answer several of the questions posed in that paper and to pose some new ones. The notation and terminology of this paper will adhere as much as possible to accepted standards but some of the main points are listed here. The relation a b means that | a \ b | < 0 while f g means that f and g belong to or, perhaps, A where A is innite and f (n) g(n) for all but nitely many integers n. If f (n) < g(n) for all but nitely many integers n then this will be denoted by f < g. By a chain in will be meant a subset of which is well ordered by < and consists of nondecreasing functions. In the next section the eects of modifying this denition of a chain will be discussed. A subset S will be said to be unbounded if for every f there is g S such that g f . The least cardinality of an unbounded subset of is denoted by b while the least cardinality of a conal subset of is denoted by d. The term ultralter will be reserved for ultralters on which contain no nite sets. A P -point is an ultralter on , U , such that for every A [U ] there is B U such that B A for every A A. If U is a lter then U will denote the dual ideal to U .
The rst author is partially supported by the basic reasearch fund of the Israeli Academy. The second author is partially supported by NSERC and was a guest of Rutgers University while the research on this paper was being done. The authors would also like to thank P. Nyikos for his valuable comments on early versions of this paper. This is number 465 on the rst authors list of publications.
1

465

revision:1993-08-24

modified:2002-07-16

SAHARON SHELAH AND JURIS STEPRANS

If U is an ultralter then the integers modulo U will refer to the ultrapower of the integers with respect to U and will be denoted by /U . If U is an ultralter then C will be said to be unbounded modulo U if, letting [f ]U represent the equivalence class of f in /U , the set {[f ]U : f C} is unbounded in the linear order /U . The least ordinal which can be embedded conally in a linear ordering L is denoted by cof(L) cof( /U ) will be an important invariant of U in the following discussion. For reference, here are Nyikos axioms (throughout C refers to a maximal chain of nondescending functions in and U refers to an ultralter) Axiom Axiom Axiom Axiom Axiom Axiom Axiom Axiom 1 2 3 4 5 5.5 6 6.5 (U )(C)(C is unbounded modulo (C)(U )(C is unbounded modulo (U )(C)(C is unbounded modulo (U )(C)(C is unbounded modulo (C)(U )(C is unbounded modulo (U )(cof( /U ) = b) (C)(U )(C is unbounded modulo (C)(U )(cof(C) = cof( /U )) U) U) U) U) U) U)

Various implications and non-implications between these axioms are established in [5]. As well, it is observed that Axiom 2 is equivalent to the equality b = d.
modified:2002-07-16

2. Non-monotone functions The denition of chains as < -increasing sequences of nondecreasing functions in the axioms which appeared in [5] may appear to be somewhat arbitrary and one may wonder what results if chains are dened dierently. For the record, therefore, the following denitions are oered. Denition 2.1. If A is a subset of then by a (< , A)-chain will be meant a subset of A which is well ordered by < . By a ( , A)chain will be meant a subset of A which is well ordered by . For the purposes of this denition the most important subsets of are: the nonodecreasing functions, which will be denoted by N , the strictly increasing functions, which will be denoted by S, and . If x {< , } then Axiom N (x, A) will denote the Axiom N with C being a variable ranging over (x, A)-chains so Axiom N is the same as Axiom N (< , N ).

465

revision:1993-08-24

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

Fortunately, many of these axioms turn out to be equivalent and others are simply false. The following simple observation of Nyikos can be used to see this. Lemma 2.1. There is a mapping : such that (f ) is strictly increasing for every f f (f ) for every f if f g and f = g then (f ) < (g) if f is nondecreasing and f (n) < g(n) then (f )(n) < g(n)(n + 1) + n Proof: Dene (f )(n) = ( n f (i)) + n. It is easy to check that i=0 f (f ) and that (f ) is strictly increasing. If f g and f = g then there is some m such that f (i) g(i) for all i m. Since there are innitely many k such that f (i) = g(i) is follows that there is some M > m such that M g(i) > m (f (i) g(i)). Hence i=0 i=0 (f )(j) < (g)(j) for all j M . Finally observe that if f (n) < g(n) and f is nondecreasing then f (i) < g(n) for each i n. Hence (f )(n) f (n)(n + 1) + n < g(n)(n + 1) + n. A consequence of Lemma 2.1 is that given any ( , )-chain {f : } there is a (< , S)-chain {f : } such that f f for each . Consequently, Axiom N (< , S), Axiom N (< , N ), Axiom N (< , ), Axiom N ( , ), Axiom N ( , S) and Axiom N ( , N ) are all equivalent for N {2, 4, 6, 6.5}. Therefore, from now, if N {2, 4, 6, 6.5}, then Axiom N will be used to denote any and all of the Axioms N (x, A) where x {< , }, A {N , S, }. Another consequence of Lemma 2.1 is that given any ( , N )-chain {f : } there is a (< , S)-chain {f : } such that f f for each and, for each ultralter U and g , the function g is an upper bound for {f : } modulo U if and only if g (n + 1) + n is an upper bound for {f : } modulo U . Consequently, Axiom N (< , S), Axiom N (< , N ), Axiom N ( , S) amd Axiom N ( , N ) are all equivalent for N {1, 3, 5}. Moreover, Axiom 3(< , ) and Axiom 3( , ) are obviously both false because if U is any ultralter then it is possible to choose X such that X U / and then nd, using Lemma 2.1, a (< , )-chain C such that f (n) = 0 for f C and n X. It follows from this that Axioms 1(< , ) and 1( , ) are also false. Therefore, from now on Axiom N can be used to denote any and all of the Axioms N (x, A) where x {< , }, A {N , S, } and N {1, 3}. Also the notation Axiom 5(N ) can

465

revision:1993-08-24

modified:2002-07-16

SAHARON SHELAH AND JURIS STEPRANS

be used to denote any and all of Axiom 5(< , S), Axiom 5(< , N ), Axiom 5( , S) amd Axiom 5( , N ). It is worth noting that Axiom 5(N ) is not equivalent to Axiom 5(< , ) or Axiom 5( , ). The reason for this is that it will been shown, in Theorem 2.4, that Axiom 2 implies Axiom 5(N ); but the same is not true of Axiom 5(< , ) since the next result shows that Axiom 5(< , ) fails assuming 20 = 1 . It is obvious that Axiom 2 holds if 20 = 1 . The following denition will be used to establish this and appears to be central in the context of non-monotone functions. Denition 2.2. If A then dene I (A) to be the set of all X such that {f X : f A} is bounded. Notice that I (A) is an ideal and that I (A) is proper if and only if A is an unbounded subset of . It is also worth observing that if C and U is an ultralter and C is unbounded modulo U then U I (C) = . Lemma 2.2. If there is a sequence {X : 1 } of subsets of such that X X if X \ X is innite if there exists a family {g : 1 } such that for every f there is 1 such that g X+1 \ X f X+1 \ X then there is an unbounded (< , )-chain, C, such that I (C) contains {{n : f (n) n} : f C}. Proof: Let {X : 1 } and {g : 1 } satisfy the hypothesis of the lemma and, without loss of generality, assume that g (n) n for all and n. Let {h : 1 } be a < -increasing sequence of functions such that h (n) < n for all n . A standard induction argument can now be used to construct {f : 1 } such that f \ X+1 = h \ X+1 if then f f if then f X = f X f X+1 \ X = g X+1 \ X and this clearly suces. Notice that 1 is crucial to the proof of Lemma 2.2 and can not be replaced by a larger cardinal. The reason is that the inductive construction relies on the fact that if {fn : n } is a family of partial functions from to such that fn = fn+1 dom(fn ) then, there is a single function f such that fn f for each n . A Hausdro gap

465

revision:1993-08-24

modified:2002-07-16

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

type of construction shows that this is not possible if 1 is replaced by some larger cardinal. It is for the same reason that 1 appears in the next corollary. Corollary 2.1. If d = 1 then there is an unbounded (< , )-chain, C, such that I (C) contains {{n : f (n) n} : f C}. Proof: Let {X : 1 } be any sequence of subsets of such that X X X \ X is innite if . Then {g : 1 } can be any dominating family. The next result shows that if d = 1 then Axiom 5(< , ) fails. Theorem 2.1. If d = 1 then there is an unbounded (< , )-chain which is bounded modulo any ultralter. Proof: Use Corollary 2.1 to nd an unbounded chain C such that I (C) contains {{n : f (n) n} : f C}. If V is any ultralter such that C is unbounded modulo V then it must be that V {{\ : {(\) \} : { C} = . Then, it is clear that the identity function is an upper bound for C. It seems that Axiom 5(< , ) is very strong and Axiom 5( , ) is potentially even stronger. Nevertheless, Axiom 5( , ) is consistent and does not imply Axiom 1. This is implied by the next sequence of results. The question of which of the axioms are implied by Axiom 5( , ) is mostly open however. The Open Colouring Axiom was rst considered by Abraham, Rubin and Shelah in [1] and later strengthened by Todorcevic [8]. Denition 2.3. The Open Colouring Axiom states that if X R and V [X ] is an open set1 then either there is Y [X]1 such that [Y ]2 V or there exists a partition of X = n Xn such that [Xn ]2 V = for each n . R can be replaced by any second countable space in the statement of the Open Colouring Axiom. Theorem 2.2. If the Open Colouring Axiom holds and {(h , g ) : } satises is a regular cardinal greater than 1 dom(h ) = dom(g ) = X for if then X X if n X then g (n) h (n) if then g g X h
1Here

revision:1993-08-24

modified:2002-07-16

[X]2 can be thought of as the set of points in X 2 above the diagonal.

465

SAHARON SHELAH AND JURIS STEPRANS

then there exists a function f : such that g f .

X for all

Proof: To begin, identify with the subspace of the reals {(h , g ) : } the reals are being considered as ( )2 or, in other words, the irrationals. Dene V = {{, } []2 : and (n)(g (n) > h (n))} and observe that V is open. From the Open Colouring Axiom it follows that there are only two possibilities. The rst is that there is a partition = n Xn such that [X]2 V = for each n . In this case there must be some n such that Xn is conal in . Choose such that X = X and let f (n) = min{h (n) : }. Now, if > and n X then there is some such that n X and hence n dom(f ). Moreover, if and n X then g (n) h (n) and so g (n) f (n). So f is the desired function. The second possibility is that there is X []1 such that [X]2 V . Since 2 it is possible to choose some such that X . It is then possible to choose M , g : M and Y [X]1 such that if Y then X \ M X if Y and n X \ M then g (n) g (n) h (n) g M = g Then if and {, } [Y ]2 and n X X it must be that either n M or n < M . In the rst case it follows that n X and so g (n) g (n) h (n). In the second case it may be concluded that g (n) = g(n) = g (n) h (n). It follows that {, } V which is a / contradiction. Theorem 2.3. The conjunction of Axiom 2 and the Open Colouring Axiom implies Axiom 5( , ). Proof: To begin, recall that it was shown in [5] that Axiom 2 implies that b = d. Hence it is possible to choose a ( , )-chain {d : d} which is also a dominating family in . Also, if C is any ( , )-chain then C is of the form {g : d}. Dene E(, ) = {n : g (n) d (n)}. Next, let {M : d} be a sequence of elementary submodels of (H(c+ ), ) such that | M | < d for each d {g : d} M and {d : d} M for each d M

465

revision:1993-08-24

modified:2002-07-16

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

M M for each d and let () = M d. Dene F to be the lter generated by {E((), ( + 1)) : d and is odd } and observe that if F is a proper lter then C will be conal in modulo U for any ultralter extending F . Hence it suces to show that F is proper. To this end let F be the lter generated by {E((), ( + 1)) : and is odd } and prove by induction that each F is proper. Moreover, it will be shown by induction that F I (C) = . If = 0, is odd or is a limit then there is nothing to do so suppose that = + 1, where is odd, and that F is a proper lter such that F I (C) = . Notice that F M because is odd. Therefore it suces to show that for each B I (C)+ there is some d such that E(( ), ) B I (C)+ the reason being that the elementarity of M +1 will guarantee that E(( ), ( + 1)) B I (C)+ for each B I (C)+ . Elementarity also assures that it may as well be assumed that B M . But if there is some B I (C)+ such that E(( ), ) B I (C) for each c then it is possible to nd h such that dom(h ) = B E(( ), ) for each d h (n) g (n) for every n B E(( ), ) and for each d g dom(h ) h for each c It follows that {(h , g BE(( ), )) : d} satises the hypothesis of Lemma 2.2. Since the Open Colouring Axiom is being assumed, there is a function f such that g f B E(( ), ) for each d. It follows that for each d there are only nitely many n B such that g (n) > max{d (n), f (n)} contradicting that B I (C). / Notice that it is shown in [8] that the Proper Forcing Axiom implies the hypothesis of Theorem 2.3. Moreover it is a Corollary that Axiom 5( , ) does not imply Axiom 1 because it is easy to check that Martins Axiom and hence the Proper Forcing Axiom implies that Axiom 1 fails. In particular, it is possible to inductively dene a (< , S)-chain no member of which dominates the exponential2 function. It has already been mentioned that the next lemma can be used to show that Axiom 5(N ) is not equivalent to Axiom 5( , ) or
exponential function is not crucial here but some quickly growing function must be used. For example, although the identity function is strictly increasing it can not be used because it is the minimal strictly increasing function.
2The

465

revision:1993-08-24

modified:2002-07-16

SAHARON SHELAH AND JURIS STEPRANS

Axiom 5(< , ). It will also be used in the proof of Theorem 3.2 but also has some interest on its own since it provides a sucient condition for Axiom 5(N ) to hold. Thus, it will be used to show that Axiom 2 implies Axiom 5(N ). Lemma 2.3. If d is regular and {c : d} is increasing and, moreover, {c A : d} is unbounded in A for each A []0 then there is an ultralter U such that {c : d} is conal in /U . Proof: Let D be a conal family in of cardinality d. Let {M : d} be an increasing sequence of elementary submodels of (H(c+ ), {c : d}, D, ) such that M d = () c for each c and d M D this is possible because d is regular. Let U = {{\ : {(\)
() (\)}

: d and f M }

and note that it suces to show that this is a base for a lter. That U has the nite intersection property can be established by induction. Let B(, f ) = {n : f (n) c() (n)} for c and f M and suppose that | A | = for each A [U ] the case m = 1 is an easy consequence of elementarity. Now let {B(0 , f0 ), B(1 , f1 ), . . . , B(m , fm )} [U ]
modified:2002-07-16

and suppose that i i+1 for each i. If m1 = m then {fm1 , fm } Mm and so the elementarity of Mm ensures that there is some g Mm such that fmj g for each j 2. Hence B(0 , f0 )B(1 , f1 ) . . . B(m , fm ) contains B(0 , f0 ) B(1 , f1 ) B(m2 , fm2 ) B(m , g) and this set is innite by the induction hypothesis. On the other hand, if m1 m then B = B(0 , f0 ) B(1 , f1 ) B(m1 , fm1 ) is innite by the induction hypothesis and, moreover, B belongs to Mm because all the parameters dening it do. Since {c B : d} is unbounded in B it follows that there must be some Mm such that fm B c B and so fm B c() B. Since B(0 , f0 ) B(1 , f1 ) B(m , fm ) = B B(m , fm ) this is enough.

revision:1993-08-24

Theorem 2.4. Axiom 2 implies Axiom 5(N ).

465

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

Proof: In [5] it is shown that Axiom 2 is equivalent to the equality b = d. Since b is regular it follows that d is regular. Moreover, if C is an unbounded ( , N )-chain then cof(C) = d. Since C consists of nondecreasing functions it is clear that {c A : c C} is unbounded for each innite set A. Hence, by Lemma 2.3, it follows that there is an ultralter U such that C is unbounded modulo U .

3. Oracle Chain Conditions and Locally Cohen Partial Orders It will be shown that there is a model of set theory where Axiom 6.5 fails. This answers the rst two questions in Problem 5 of [5]. C. Laamme has remarked that in some models of NCF (see [2] for an overview of this area) Axiom 6.5 fails as well because it is possible to provide a classication of chains in these models. The restriction to chains does not play an important role in this theorem and, in fact, the theorem is slightly stronger than required at least formally because of this. Theorem 3.1. There is a model where cof( /U ) = for every ultralter U but every unbounded subset of has an unbounded subset of size 1 .
modified:2002-07-16

465

revision:1993-08-24

Proof: The plan of the proof is to start with a model V in which 1 and 2 (1 ) in other words, the trapping of subsets of 2 occurs at ordinals of conality 1 in 2 both hold. In this model a nite support iteration {(P , Q ) : 2 } will be constructed along with a sequence of oracles [6] {M : 2 } more precisely, M is a P name for an oracle. The oracles will be chosen so that if {g : 1 } is a P -name, guessed by the 2 (1 ) sequence, for an unbounded subset of then M is chosen so that if Q is any partial order satisfying the M -chain condition then forcing with Q does not destroy the unboundedness of {g : 1 }. Provided that P2 /P satises the M -chain condition, it will follow that every unbounded subset of has conality 1 because every unbounded subset is reected at some initial stage by the 2 (1 ) sequence. The rest of the result will follow once it is shown how to construct Q satisfying the M -chain condition and adding an upper bound to any given sequence from some ultrapower of the integers. The construction of {(P , Q ) : 2 } and {M : 2 } is, of course, done by induction. If is a limit then P is simply the direct

10

SAHARON SHELAH AND JURIS STEPRANS

465

revision:1993-08-24

limit of {P : }. The construction of M and Q does not depend on whether or not is a limit. Given P , use the results of pages 124 to 127 of [6] to nd a P -name for a single oracle N such that if Q saties the N-chain condition then it satises the M chain condition for each . Let C be the set guessed by the 2 (1 ) sequence at . If C is not a P -name for an unbounded subset of then let M = N . Otherwise, use Lemma 2.1 on page 122 of [6] to nd an oracle M such that if Q satises the M-chain condition then the subset C remains unbounded after forcing with Q. The use of Lemma 2.1 requires checking that if C is an unbounded chain then adding a Cohen real will not destroy its unboundedness. This is a result of the folklore which can be found in [7]. Then use the results of pages 124 to 127 of [6] to nd a single oracle M such that any Q which satises the M -chain condition will also satisfy the M-chain condition and the N -chain condition. Suppose that the 2 (1 ) sequence has also trapped a lter U which is an ultralter in the intermediate generic extension by P and an increasing sequence {f : 1 } in the reduced power of the integers modulo U . (So it is being assumed that, by some coding, the 2 (1 ) sequence traps triples of sets the rst component of the triple at is a candidate for C in the construction of M while the second and third components are candidates for the ultralter U and the sequence {f : 1 }.) The only thing left to do is to construct Q satisfying the M -chain condition and adding an upper bound for {f : 1 } in the reduced power modulo U . Let M = {M : 1 }. The partial order Q is constructed by induction on 1 in V P it will be similar to the forcing which adds a dominating real but with extra side conditions. In particular, a sequence of partial functions {S : 1 } is constructed by induction on 1 and Q is dened to be the set of all pairs (F, ) such that F : k is a nite partial function and []<0 . The ordering on Q is dened by (F, ) (F , ) provided that , F F and F (j) S (j) for and j (dom(S ) \ dom(F )). Moreover, the functions S will be constructed so that dom(S ) U and S (j) f (j) for each j dom(S ). It is easy to see that {(F, ) : } is dense in Q for every and so if G is Q = Q1 generic over V P then {F : ()((F, ) G)} is an upper bound for {f : 1 } in the reduced power with respect to U . It therefore suces to construct {S : 1 } so that for every 1 , every dense open subset of Q

modified:2002-07-16

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

11

modified:2002-07-16

which belongs to M remains predense in Q+1 . This, of course, will ensure that Q = Q1 satises the M -chain condition. Suppose that {S : } have been constructed. Let A be the set of all of the dense open subsets of Q which belong to M this includes all those dense open subsets of Q which belong to M some . Choose h to be some function which dominates all members of M ; in other words, if g M then g h. Let {(Ai , (Fi , i )) : i } enumerate A Q . Now choose, by induction on , integers {Ki : i } such that Ki < Ki+1 and K0 = 0. Given Ki , dene Fji Fj for each j i such that if dom(Fj ) Ki then dom(Fji ) = Ki , (Fj , j ) (Fji , j ) and Fji (k) h(k) if k dom(Fji \ Fj ). Now choose (Fji , i ) Aj such that (Fji , i ) (Fji , ). Let Ki+1 be such that j j dom(Fji ) Ki+1 for each j i. Let Xm = i [K2i+m , K2i+m+1 ) for m 2 and note that there exists m 2 such that Xm U . Let S = h Xm ; the reason being that U is an ultralter in V P and X0 V P . To see that this denition of S ensures that every dense open subset of Q which belongs to M remains predense in Q+1 let (F, ) Q+1 and let D M be dense open in Q . It follows that (F, \ {}) Q . To simplify notation assume that m = 1. Choose j such that dom(F ) K2j and such that (D, (F, \ {}) = (Ak , (Fk , k )) for some k 2j. Since 2j 2j 2j dom(Fk ) K2j it follows that F (n) h(n) if n dom(F \ Fk ). k k 2j Moreover [K2j , K2j+1 ) domS = . Hence (Fk , 2j ) is compatible k with (F, ).

The methods of the previous theorem can also be used to show that it is consistent that Axiom 6 holds but Axiom 5.5 fails. In establishing this it will be helpful to introduce the following denition. Denition 3.1. A partial order (P, ) will called locally Cohen if for every X [P]0 there is Y [P]0 such that X Y and Y is completely embedded in P in other words, if A Y is a maximal antichain in the partial order (Y, Y Y ) then it is also maximal in (P, ). The notion of locally Cohen partial orders has already been isolated and investigated by W. Just in [4] who refers to locally Cohen partial orders as harmless. The motivation of Just was that any locally Cohen forcing satises the oracle chain condition for every oracle. Let S() be the canonical partial order for adding a scale of length in with nite conditions. To be precise, a condition p belongs to

465

revision:1993-08-24

12

SAHARON SHELAH AND JURIS STEPRANS

S() if and only if p : p np is a function and p []<0 and np . The ordering on S() is dened by p q if and only if: pq if {, } p and then q(, m) q(, m) for every m nq \ np It should be noted that that S() is also the nite support iteration of length of the partial orders {D() : } where D() is the nite condition forcing for adding a nondecreasing function which will be denoted by c which dominates all the reals {c : }. Lemma 3.1. For any ordinal the partial order S() is locally Cohen. Proof: Given X [P]0 let Y []0 be any set such that S(Y ) X S(Y ) can be dened for any set of ordinals in the same way that S() is dened for an ordinal . To see that S(Y ) is completely embedded in S() let A S(Y ) be a maximal antichain in S(Y ). If p S() then let = Y and p = p np . Since p S(Y ) there must be some q A such that q = p q S(Y ) and p q . Dene q : (q p ) nq by if Y q (, j) max{q (, j) : Y } if j np and Y / / q (, j) = p(, j) if Y and j np / It is easy to check that p q .
modified:2002-07-16

Theorem 3.2. There is a model of set theory where 2 0 = 2 b = 1 there is a an unbounded ( , )-chain of length 2 the conality of any ultrapower of the integers is 2 Proof: As in the proof of Theorem 3.1, let V be a model of set theory where 1 and 2 (1 ) are both satised. Let H be S(2 ) generic over V and dene H = H S(). Let c (n) = (H)(, n) and observe that {c : 2 } is increasing with respect to and {c : 2 } is not bounded. If an oracle chain condition forcing extension of V [H] can be found which preserves the unboundedness of {c : 2 } and in which the conality of any ultrapower of the integers is 2 then the result will follow because b = 1 is easily preserved by the oracle chain condition. To do this, construct {(P , Q ) : 2 } and {M : 2 } exactly as in the proof of Theorem 3.1 except that M is chosen to be an oracle

465

revision:1993-08-24

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

13

in the model V [G, H ] where G is generic over P . There is no problem in doing this because S() is locally Cohen and hence satises the M chain condition for each indeed, S() satises any oracle chain condition. It is therefore easy to use Claim 3.3 on page 127 of [6] to obtain M exactly as in the proof of Theorem 3.1. Let G be P2 generic over V [H]. Exactly as in the proof of Theorem 3.1, it can be shown that the conality of any ultrapower of the integers is 2 while b = 1 in V [G, H]. On the other hand, the fact that {c : 2 } is unbounded follows from genericity and the fact that S( + 1) = S() D({c : }) so c is not dominated by any function from V [G , H ] where G is the restriction of G to P . Corollary 3.1. Axiom 6 does not imply Axiom 5.5. To see that the model constructed in Theorem 3.2 is a model of Axiom 6 but not of Axiom 5.5 observe rst that Axiom 5.5 fails because b = 1 while the conality of any ultrapower of the integers is 2 . On the other hand, there is ( , )-chain of length 2 d. Using Lemma 2.1 it is possible to construct from this a (< , S)-chain, C , of nondecreasing functions. From Lemma 2.3 it follows that there is an ultralter U on such that C is conal in the ultrapower of the integers modulo U . This is the statement of Axiom 6.
modified:2002-07-16

The partial order S() can be modied to yield a model where Axiom 4 holds yet Axiom 5.5 fails. Recall that Lemma 2.1 implies that to do this it is only necessary to nd a model of Axiom 4(, ) and the failure of Axiom 5.5. Theorem 3.3. If set theory is consistent then there is a model of set theory where b = 1 yet for every ultralter U there is a ( , )-chain of length 2 which is conal in /U . Proof: It will be shown that, assuming 2 (1 ), there is a locally Cohen partial order P such that if G is P generic then cof( /U ) = for every ultralter U in V [G]. The fact that P is locally Cohen will guarantee that b = 1 in V [G]. To construct P some preliminary bookkeeping is required. Let {D : 2 } be a 2 (1 ) sequence and let {g : 2 } enumerate names for elements of which arise from countable chain condition forcing partial orders on 2 . Also, if Q is any partial order of size 2 and satisfying the countable chain condition then any subset of the reals in a Q generic extension has a name of size 2 . Consequently, it is

465

revision:1993-08-24

14

SAHARON SHELAH AND JURIS STEPRANS

possible to use subsets of 2 to code such names for sets of reals. If X is some name in a suitable partial order for a subset of []0 then c(X ) will denote the subset of 2 which codes it while if X 2 then d(X) will denote the name it codes. The details of the coding will not be important. Dene a partial order on 2 by if and only if c(D ) = c(D ). Now construct P as a nite support iteration of {P : 2 } such that P+1 = P C D where C adds a Cohen real, A : 2 and D is some partial order which has yet to be dened. At the same time, construct a partial function : 2 2 so that if is in the domain of then 1 P d(D ) is an ultralter and () is the minimum ordinal such that () {() : / } and such that g () is a P name. Given P , dene D by p D if and only if p = (fp , p ) fp p []<0 if p then

and p q is dened to hold if and only if fp fq p q if p , A1 {k} d(D ), A (n) = k and n dom(fq \ fp ) then fq (n) max{g() (n), F (n)} where, for any 2 , F is the generic function added by the partial order D to be precise, F = {fp : p G} where G is D generic . If it is possible to extend to include in its domain then do so there is no ambiguity here because an extension, if it exists, is unique. Let P = P2 . It will soon be shown that P satises the countable chain condition. However, rst suppose that G is P generic over V and that U is the P name for an ultralter in V [G]. There is then a stationary set, S(U ), such that if S(U ) then 1 P d(c(U ) ) is an ultralter. It will be shown that {F : S(U )} is a ( , )chain which is conal in /U . The fact that it is a increasing sequence is an immediate consequence of the denition of D . To see that it is conal in /U let g . Then, assuming that P has the countable chain condition, there is some 2 and 2 such that g is a P name for g. It follows that there is some S(U ) such that = (). Let S(U ) \ ( + ) and note that . Hence, the partial order D adds a function which dominates g on A1 {k} for some k 2 and, moreover, A1 {k} D .

465

revision:1993-08-24

modified:2002-07-16

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

15

465

revision:1993-08-24

It remains to be shown that P satises the countable chain condition and that b = 1 after forcing with P. Both these facts will follow once it has been shown that P is locally Cohen. To this end, it is worth observing that P has a dense set of conditions which are somewhat determined a condition p will be said to be somewhat determined if the support of p is p [2 ]<0 and there is an integer n(p) such that for each p dom() there is hp, : n(p) 2, hp, : 0 1 n(p) and p, []<0 such that p P p() = (hp, , hp, , p, ) 0 1 p, { p : } for each p and p such that there is k(p, , ) 2 such that p P A1 {k(p, , )} d(D ) for each p and p such that there is Mp () and Gp () : Mp () such that p P g() Mp () = Gp () and, moreover, hp, (i) = 1 + k(p, , ) mod 2 provided 0 that i n(p) \ Mp (). The fact that the set of somewhat determined conditions in P is dense in P will be proved by induction, but an extra induction hypothesis is necessary. What will be shown by induction on is that, given p P any nite set W of maximal elements of ( ) any function v : W 2 any function a : W 2 such that ga() is a P name for each W there is a determined condition q the fact that q is determined is witnessed by n(q) with the additional properties that for each W there is M () and G() : M () such that q P ga() M () = G() and, moreover, hp, (i) = v() provided that 0 i n(q) \ M (). If = 0 this is trivial and if is a limit ordinal then it follows from the fact that a nite support iteration is being used. Therefore, suppose that the fact has been established for and that p = (p , (h0 , h1 , )) P+1 . Suppose also that W , v : W 2 and a : W 2 have been given so that W is a nite set of maximal elements of ( +1)2 . Notice that ( +1)2 has at most one maximal element, , which is not maximal in ( ). It is, of course, possible that some maximal element in ( ) is no longer maximal in 2 ( + 1) . If there is such a new non-maximal element, then denote it by ; if not, then the following argument is a bit easier and so it will be assumed that exists. Find q p and H0 : I0 2 and H1 : I1 as well as k 2 and []0 such that

modified:2002-07-16

16

SAHARON SHELAH AND JURIS STEPRANS

q P h0 = H0 and h1 = H1 q P ga() I0 = G for some G : I0 (if W this can / be ignored) q P = q P A1 {k} d(D ) That it is possible to arrange for the rst two clauses follows from the fact that ga() is a P name and so any information about it can be obtained without changing h0 or h1 . To satisfy the last clause, use the fact that , which follows because is no longer maxinal in ( + 1)2 . Now dene W = (W \ {}) {} and observe that W is a set of maximal elements in ( ). Dene v = v W {(, k + 1 mod 2)} and a = a W {(, ())} and observe that both a and v are still functions of the right type. Then use the induction hypothesis on to nd q q which is somewhat determined and such that this is witnessed by n(q ) and, such that for each W there is M () and G() : M () such that q P ga () M () = G() and, moreover, hp, (i) = v () provided that i n(q) \ M (). Without loss 0 of generality, n(q ) I0 and n(q ) I1 . Then let p = (q , (h0 , h1 , ) where h0 : n(q ) 2 is the extension of H0 to n(q ) such that h0 (i) = v() if i n(q ) \ I0 and h1 is the extension of H1 such that
modified:2002-07-16

h1 (i) = max({f (i) :

and }{G(())(i) :

and })

465

revision:1993-08-24

for i n(q ) \ I1 . Notice that maximum is taken over actual integers rather than names for integers. The denition also respects the requirements of extension in the partial order P . Dening M () = I0 and G() = G satises the extra induction hypothesis. To see that P is locally Cohen let X [P]0 . Let M be a countable elementary submodel of (H(3 ), P, {D : 2 }, , X). It suces to show that P M is completely embedded in P. To see that this is so, let A P M be a maximal antichain in P M and let p P; without loss of generality p can be assumed to be somewhat determined and, moreover, it may be assumed that this is witnessed by n(p). Let p be dened so that dom(p ) = dom(p)M and p () = (hp, , hp, , p M) 0 1 for dom(p ). Note that p M P. Hence there is q A and q P M such that q q and q p without loss of generality it may be assumed that q is determined and this is witnessed by n(q). It must be shown that p and q are compatible.

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

17

As in the proof that S() is locally Cohen, for dom(p) \ dom(q) extend hp, to h by dening 1 1 h (m) = max({hq, (m) : 1 1 . . . {Gq ( )(m) : and dom(q)} . . .

and dom(q) and m Mq ( ) and A (m) = k(q, , )})

modified:2002-07-16

for m n(q) \ n(p), recalling that Gq (), Mq () and k(q, , ) are witnesses to the fact that q is somewhat determined. This will certainly assure that if and are in the domain of q and then hq, (m) h (m) hq, (m) the fact that h (m) hq, (m) follows 1 1 1 1 1 from the dention of the third coordinates in p . Also, g( ) (m) h (m) if A (m) = k(p, , ) will be true if m M ( ) by construction. 1 Next, if dom(p) \ M, dom(p) M and then dene h (m) = 1 + k(p, , ) mod 2. If this can be done then, if m n(p), 0 it is not necessary for hq, (m) to be greater than g() (m). If there is 1 no dom(p) M such that , do not extend hp, at all. Notice 0 that in this last case it is still possible that there is some dom(q) such that . However, because it is only necessary for hq, (m) to 1 be greater than g() (m) in case q, , this will cause no problems because q, M if M. What must be checked, though, is that no conict arises as a result of this denition of h . After all, it is conceivable that and 0 but k(q, , ) = k(q, , ). To see that this does not happen, suppose that , , k(p, , ) = k(p, , ) and {, } M. It follows that if = sup{ : and and dom()}

then M. Hence and so there is some such that , and dom() such that . Hence A1 {0} is measured by the ultralter d(D ). Since d(D ) d(D ) and d(D ) d(D ) it follows that k(p, , ) = k(p, , ).

revision:1993-08-24

4. Open Questions Table 1 of implications and non-implications summarizes the known results about the axioms discussed in this paper. The key to understanding Table 1 is that if there is a in the entry in the row headed by Axiom R and the column headed by Axiom C then Axiom R implies Axiom C

465

18

SAHARON SHELAH AND JURIS STEPRANS

Table 1. Table of Implications Axiom 1 2 3 4 5 5< 5N 5.5 6 6.5 1 2 ? ? 3 ? ? 4 5 5< 5N 5.5 6 6.5 ? ? ? ? ? ? ?

Table 2. Table of References Axiom 1 2 3 4 5 5< 5N 5.5 6 6.5 1 T N N N 10 10 N 5 5 5 2 T T N N ? ? N 5 5 5 3 T 1 T N ? ? N 5 5 5 4 5 5< 5N 5.5 6 6.5 T ? ? T N T N T 1 1 2 N T N N 9 9 T N T N T 3 3 T 4 T N ? T T T N T N ? ? T T N T N N 9 9 T N T N 5 9 9 ? T N N 5 6 6 8 7 T N 5 6 6 5 5 ? T

modified:2002-07-16

revision:1993-08-24

if there is a in the entry in the row headed by Axiom R and the column headed by Axiom C then Axiom R is known to be consistent with the negation of Axiom C if there is a question mark in the entry in the row headed by Axiom R and the column headed by Axiom C then it is not known whether Axiom R implies Axiom C Not all the reasons for the assertions made in Table 1 are contained in in this paper. Some will be found in in [5] and others must be deduced by modus ponens. Table 2 contains a guide to reasons for the various assertions in Table 1. A T in the row corresponding to Axiom R and the column corresponding to Axiom C in Table 2 indicates that the fact that Axiom R

465

MAXIMAL CHAINS IN

AND ULTRAPOWERS OF THE INTEGERS

19

implies Axiom C is a trivial implication by trivial is meant something which can be deduced by considering the quantiers in the relevant axioms. An N in that entry means that either the implication or non-implication can be found in [5]. The enumeration of the following list corresponds to the numbered entries in Table 2. So, for example, the second entry of this list refers to Theorem 2.4 because this is the reason there is an in Table 1 in the row corresponding to Axiom 2 and the column corresponding to Axiom 5(N ). (1) Theorem 2.1 (2) Theorem 2.4 (3) The fact that Axiom 4 does not imply Axiom 5(< , ) follows because it has been shown that Axiom 2 does not imply Axiom 5(< , ) in Theorem 2.1 and the fact that Axiom 2 implies Axiom 4 follows from an inspection of the quantiers involved. Modus ponens yields the rest. (4) Theorem 3.3 (5) The antecedent of the implication is implied by Axiom 5(N ) so the non-implication follows from modus ponens because Axiom 5(N ) does not imply the conclusion. (6) The antecedent of the implication is implied by Axiom 4 so the non-implication follows from modus ponens. (7) Corollary 3.1 (8) Axiom 5(N ) implies Axiom 5.5. (9) 20 = 1 is known [3] to imply Axiom 3 and Lemma 2.2 shows that Axiom 5(< , ) fails under this assumption. For the rest, use modus ponens. (10) See the remarks following Theorem 2.3. References
1. A. Abraham, M. Rubin, and S. Shelah, On the consistency of some partition theorems for continuous colorings, and structure of 1 -dense real order types, Ann. Pure Appl. Logic 29 (1985), 123206. 2. A. Blass, Applications of superperfect forcing and its relatives, Set Theory and its Applications (Berlin) (J. Steprns and S. Watson, eds.), Lecture Notes in a Mathematics, vol. 1401, Springer-Verlag, 1989, pp. 1840. 3. Alan Dow, Remote points in spaces with -weight 1 , Fund. Math. 74 (1984), 197205. 4. W. Just, A modication of Shelahs oracle-c.c. with applications, Trans. Amer. Math. Soc. 321 (1990), no. 2, 621645. 5. P. Nyikos, Special ultralters and conal subsets of , preprint. 6. S. Shelah, Proper forcing, Lecture Notes in Mathematics, vol. 940, SpringerVerlag, Berlin, 1982.

465

revision:1993-08-24

modified:2002-07-16

20

SAHARON SHELAH AND JURIS STEPRANS

7. J. Steprns, Combinatorial consequences of adding Cohen reals, Submitted to a Proceedings of the 1991 Jerusalem Conference Set Theory of the Reals, 19. 8. S. Todorcevic, Partition problems in topology, vol. 84, American Mathematical Society, Providence, 1989. Institute of Mathematics, Hebrew University, Jerusalem, Givat Ram, Israel and Department of Mathematics, Rutgers University, New Brunswick, New Jersey Department of Mathematics, York University, 4700 Keele Street, North York, Ontario, Canada M3J 1P3

465

revision:1993-08-24

modified:2002-07-16